agentic_agents 0.1.1 copy "agentic_agents: ^0.1.1" to clipboard
agentic_agents: ^0.1.1 copied to clipboard

A bounded tool-calling agent loop with budgets, sessions and streaming, plus a planner/executor and multi-agent delegation. Budgets are required, not optional.

Changelog #

0.1.1 #

  • Shortened the package description to the 60-180 character window pana scores against. Search engines truncate anything longer, so the ten points it withheld were pointing at a real defect: the useful half of the sentence was never being shown.

0.1.0 #

Initial release of the agent layer.

Added #

  • ContractAgent, AgentInfo, AgentInput, AgentResult, AgentStep and AgentStopReason. Failures return as a result carrying the step trail rather than throwing; ensureSuccess rethrows for callers who want the exception. Cancellation propagates.
  • BudgetsAgentBudget and BudgetTracker bound a run by iterations, tokens, cost, wall clock and tool calls. On the last permitted iteration the loop forbids tool calling so the run still produces an answer, and an exhausted budget explains itself in text that can be shown to a user.
  • The loopToolCallingAgent, with streaming, early-stop conditions, per-run budget overrides, tracing and events.
  • SessionsAgentSession with HistoryStrategy, KeepAllHistory, SlidingWindowHistory and CharacterBudgetHistory. Trimming repairs dangling tool results, which providers otherwise reject. Sessions serialise for persistence.
  • StreamingAgentChunk and its variants, extensible so that agent types the framework does not ship can report updates of their own.
  • DelegationAgentTool presents an agent as a tool, with depth bounding through the run context; supervisorOver builds a supervisor over a team.
  • PlanningPlannerExecutorAgent, Plan and PlanStep, decomposing a request into bounded steps that each run with a clean context.
  • EventsAgentRunStarted, AgentStepCompleted, AgentRunCompleted, AgentBudgetExhausted and AgentDelegated.

Notes #

HistoryStrategy.select is asynchronous and receives the turn about to be sent. Both were needed the moment a real memory backend was built against the seam: summarising and retrieving are asynchronous, and recall must be driven by the question just asked rather than by the previous one — a synchronous, history-only signature recalls nothing at all on the first turn.

2
likes
160
points
170
downloads

Documentation

Documentation
API reference

Publisher

unverified uploader

Weekly Downloads

A bounded tool-calling agent loop with budgets, sessions and streaming, plus a planner/executor and multi-agent delegation. Budgets are required, not optional.

Repository (GitHub)
View/report issues
Contributing

Topics

#ai #agents #llm #agentic #multi-agent

License

MIT (license)

Dependencies

agentic_core, agentic_llm, agentic_tools, meta

More

Packages that depend on agentic_agents